Thank you for this info. It does help illustrate the issue. However, it does not, IMO, make the chip makers any less guilty of deliberate fraud.
The average consumer cannot be presumed to have expert knowledge. The chip & gear makers know that a consumer will automatically assume that a dac/player claiming 32bit is better than one claiming 24bit. The fact that 99.99% of audio systems don't have even 16bits of dynamic range is irrelevant, as is the fact that 99.99% of consumers have no idea this is the case.
I guess the motivation for my thread is just that I can't understand why this deliberately deceptive, blatantly dishonest marketing of "32 bit dac chips" doesn't seem to bother anyone. Some of the respondents here give the impression that you have nothing but admiration for these chip makers, and that they don't deserve one unkind word. I don't get that attitude at all. A whole group of chip makers, with willing collusion of a whole bunch of gear makers, is comitting deliberately deceptive, if not legally fraudulent, certainly morally fraudulent business practice, and they get nothing but praise from even those of you with expert knowledge? I don't get it.
I know I'm tilting at windmills with this thread, but I felt it's an issue that deserves attention and wasn't getting any.
I blame both. The chip makers, at least they marketing minions, knew darn well that the gear makers would love to be able to sell "32bit" gear, knowing that they could simply shift blame to the chip makers if anyone called them on it.
ESS was deviously clever enough to make things doubly tempting for the gear makers. A chip that allowed them to sell "32bit" AND had everything on one chip, even the input receiver. How much more tempting can it get to a gear maker?
ESS (and TI and the others) would be deviously clever fraudsters if they didn't bring anything of value with those 32bits chipset. But they actually do bring value. They enable gear makers to more easily use 32bits data in the digital domain, which has some advantages over 24bits (in the digital domain only, for dsp, volume control, etc).
It's just very convenient to call these chipsets 32bits ones. While it can be misrepresented by marketing, it's not wrong per se.
From an engineering point of view, it would be incredibly stupid to let digital quantization noise dominate the noise floor. Adding a bit or two on the digital side costs much less current and chip area than reducing the analogue noise by 6 or 12 dB. (On top of that, quantization noise is not real additive noise but rather a weird kind of distortion, although it can be made to sound like noise with proper dithering.)
Hence, any well-designed high-dynamic-range DAC will have an input wordlength considerably longer than the SNR divided by 6 dB, although you could still argue that 32 bit is somewhat overdone.
Hence, any well-designed high-dynamic-range DAC will have an input wordlength considerably longer than the SNR divided by 6 dB, although you could still argue that 32 bit is somewhat overdone.
Plenty of reasons why it doesn't bother me. Perhaps the main one is I've learned not to trust 'headline' stuff in datasheets but always dig deeper down to see what's in the detail. After all, that's where the devil is. I came across another example where the headline for an opamp is kind of misleading, though of course not outright wrong - that's the case of the OPA1652 being billed as 'FET input'. When you drill down into the detail (most particularly the schematic is shown) you find the whole device is CMOS. Over on a headphone thread recently a member who works for TI offered some background to this, saying that marketing didn't want to mention CMOS upfront because it has a bad rap with audiophiles.
My tentative conclusion is that its not necessarily the conscious intention to mislead, rather fear of missing out in sales which is the underlying motivation not to state the unvarnished, warts-and-all truth in datasheet headlines.
Forget 32-bit DACs, they're already obsolete.
Amazon.com: 64 bit - Internal Sound Cards / Computer Components: Electronics
Amazon.com: 64 bit - Internal Sound Cards / Computer Components: Electronics
Uneducated consumers will assume anything ANYWAY to think that one stuff is better than the other stuff. Heavy stuff is better than light one, expensive one is better than cheap one, etc....
Think about the amplification power. Consumers always want BIG power. Some producers think that their Sales/Marketing strategy should address the issue. So they stick a 1000W label on the brochure. It is not achievable but it is not a lie either. Then there are other competitors who think that they will go down if they don't play by the same strategy...
Naive is probably the right key word. There is no big companies that don't do fraudulent business practice. There is probably no president that have good morale. You have to be a psychopath to win the competition to become a leader. No? Then you either have to open your eyes wider or set a different standard.
I'm amused by this review of an "8000 watt amplifier" (plugs into a USA-standard 120V outlet). The reviewer knows the rating is a lie, but he still gives it a 4 star review.
Amazon.com: MuzikMan's review of PYLE PT8000CH Rack Mount 8000-Watt 8 Chann...
Amazon.com: MuzikMan's review of PYLE PT8000CH Rack Mount 8000-Watt 8 Chann...
I had an unrelated, but similar issue with the advertising of "Gigabit Ethernet". I purchased a Buffalo brand Gigiabit NAS and when I found it running way slower than advertised I came to find out that all you need to qualify to use the term "Gigabit" is to be faster than 100 MBps. So when I was running at 101Mbps that qualified as gigabit.
As for the use of 32 bit in DAC chips, I wonder if its a similar issue. Is this not referring to the Dac's ability to directly handle files recorded in a 32 bit floating point format? I was reading an article online that shed some light on the subject for me. I am a daily user of Pro Tools and wanted to look further into why I would record in 32bit floating point over 24 bit.
https://ask.audio/articles/pro-tools-10-the-benefits-of-32-bit-floating-point-audio
It seems to me that the benefit of 32bit is basically for the DSP to have more headroom. I also spoke with a rep at Native Instruments that told me why they write all their software using 32 bit floating point. He said that it gives them a lot more flexibility to process the audio and less (almost no) chance of peaking out the signal when manipulating it.
As for the use of 32 bit in DAC chips, I wonder if its a similar issue. Is this not referring to the Dac's ability to directly handle files recorded in a 32 bit floating point format? I was reading an article online that shed some light on the subject for me. I am a daily user of Pro Tools and wanted to look further into why I would record in 32bit floating point over 24 bit.
https://ask.audio/articles/pro-tools-10-the-benefits-of-32-bit-floating-point-audio
It seems to me that the benefit of 32bit is basically for the DSP to have more headroom. I also spoke with a rep at Native Instruments that told me why they write all their software using 32 bit floating point. He said that it gives them a lot more flexibility to process the audio and less (almost no) chance of peaking out the signal when manipulating it.
Resolution is determined by several things. You cannot improve on the input, but you can improve processing accuracy so the output is not much worse than the input.
That's why, as an input, bit rate (length of 1/0 to represent signal output) is the most important, followed by sampling rate (how many samples are taken per time interval of music). Other variables are less important as they are mainly to assist in accuracy during rounding.
Before you create the source for CDP or DVP, you have ADC chip with its limited bitrate capability. Then you maintain the accuracy by using accurate 32bit software and probably 64bit computer during operation in studio. But at the end you have to create an output for the DAC input.
Input to DAC is already defined by standards. For example 16bit of bit rate and 44k1 of sampling rate for CD (Red book). DVD audio has better resolution (24bit). This is the INPUT. You cannot "improve" the output to be better than that (the theoretical accuracy).
Degradation, however, can be prevented. There are many terminologies/processes related to this. Oversampling is the most important.
32bit capability of a DAC will be critical in the future. But probably not today where the bottleneck is still somewhere else.
Okay, I don't necessarily disagree with you, just contributing to the discussion. This seems to be a good place to discuss what those chips and DAC's should be called then. As a community of professionals, hobbyists, and 'prosumers', I bet we could come up with something better. Maybe something like "+24bit" or "32Pro" (Pro for processing). I think the marketing guys would like the latter. "Buy the new New HD, 24bit, 96k, 32Pro DAC! Now with more special sauce!"
Since these DACs are capable of more than a typical 24 bit chip, I think the name should be different. I doubt TI or anyone will change their specs or anything, but at least maybe we DIY'ers can refer to them more accurately.
Since these DACs are capable of more than a typical 24 bit chip, I think the name should be different. I doubt TI or anyone will change their specs or anything, but at least maybe we DIY'ers can refer to them more accurately.
What is wrong with "32-bit DAC"? Read the full spec. Does it say it is capable of 32x6=192db?? Or worse, that it has 192db of DNR or SNR?
Remember, it is a chip, so basically it only has theoretical value. But most specification has measurement where the chip has been implemented in the real world PCB (with given parameters). There you can see what the chip is capable of. I think have never seen some kind of "lies" there.
Can anyone show me a PDF where there is a lie there? It could be just readers' interpretation.
"Lies" are probably with DAC machine producers, which is very common. I myself prefer the producer to mention the DAC chip used instead of mentioning "unsupported" specs.
Remember, it is a chip, so basically it only has theoretical value. But most specification has measurement where the chip has been implemented in the real world PCB (with given parameters). There you can see what the chip is capable of. I think have never seen some kind of "lies" there.
Can anyone show me a PDF where there is a lie there? It could be just readers' interpretation.
"Lies" are probably with DAC machine producers, which is very common. I myself prefer the producer to mention the DAC chip used instead of mentioning "unsupported" specs.
I don't know about cars so I don't know if the analogy is correct or not.
32-bit DAC could have a purpose/benefit, depends on the implementation of the chip by hardware manufacturers, especially in DSP stage, between transport that produces EFM signal and DAC that accept PCM. There are bit manipulation possibilities that will make 32-bit beneficial. But of course, it is not the major determinant of sound quality. Because first, major quality depends on the bit depth of the input. Second, the bottleneck is in chip manufacturing technology.
When I first saw "32-bit" word, I had no feeling that it should be better than 24-bit DAC in general, so I have had no problem with that...
But if "32-bit DAC" is a crime, so what, Marketing people are criminals, fine. It has been like that for decades. The DAC and the cylinder are not the only available examples.
The lie/fraud is right in the top heading of EVERY datasheet I've seen for these chips, i.e., they all are titled "32bit DAC" without qualification, and much to their DIS-credit, dac & player makers use this lie to get away with their own lies of marketing/selling "32bit" dac/player units. As I've said, NO average consumer can be expected to know that the specs reveal nowhere near 32bit actual performance.
And, again, the fact that the chip makers have *not* put on the market ANY beyond-24bit-advertised *dac-only* chip is, to me, convincing evidence that they are deliberately providing a fraudulent basis for gear makers to defraud buyers.
I just downloaded the pdf for PCM1795. In the electrical characteristics it was mentioned that the audio data input was 32-bit, and that the result was 123dB of SNR with NE5534 output stage. (123dB is around the maximum possible I guess).
My questions were: (1) What is this 32-bit audio data? (2) How much the 32-bit operation will affect the performance of 24-bit audio data input. Is the result also 123dB or what.
So, I checked with another manufacturer, the ESS, and it was clear that the DAC was intended for 32-bit audio data, such as this:
"The Sabre32 Reference DAC is a perfect companion for the emerging 32-bit audio content generated by music studios and DSP up-conversion."
So that's it. The input was not from a disc (tho blue-ray might be possible, I'm not sure), most probably computer based such as a DAW (Digital Audio Workstation) which commonly produce 32-bit audio in studios.
I think you need to read the message from SonicFreq post regarding the 32-bit floating point aritmethic operation in Pro Tools. It is not only about the Delta-Sigma.
Attachments
Bits bits bits, lost in the noise....
My little project needs 14bits (MIDI to Moog CV/Gate)
Arduino analog writes are only 8 bit PWM at 490Hz.
Dig down into the registers, maybe 12 bits...
So, the thought occurs to me: Use two PWMs and scale
them with resistors. Except the brain dead obvious 256x
ratio is so wide, its not easy to accurately dial in.
But MIDI control is only a pair of 7bits, so only 14bit
DAC is needed for those CV's. And 88 piano keys of
tuning is 186 DAC steps per key, 14bits seems plenty
there too...
So, I figure needing ony 7 bits PWM per analog write,
and a much more reasonable scaling factor of 64x. I
can switch back and forth between the redundant bits
that are within the range of either DAC to tweak 64x.
Now the tricky part. I need to low pass filter 490Hz,
and smooth the most significant bit enough that the
least significant bit isn't lost in the noise. -39dB.
I think a Sallen Key with 2 poles and a knee at 33Hz
can exceed that by a good amount, and 0.05 seconds
portamento to settle reasonably close to a new value.
Looking at 22K for both resistors, and 0.22uF for both
caps. Q is 0.5. I can't tell exact attenuation of 490Hz
from the crude graph, but sure looks to be more than
below the -40dB line...
64x scaling resistor is 1.5Meg in series with 68K.
I need 5K 25 turn pot (set to 2.5K) in series with
my Sallen's first 22K. I think that gives me enough
adjust to get away with worst case 5% resistors.
I'm never going to squeeze 7.33 octaves out of 5V
suppply rail at 1V per octave. But with 9V rail and
if I scale the Sallen's negative feedback, say 47K/
(22K+47K) I can stretch almost that far. Q rises,
but not so much its threatening to oscillate. And
gliding faster between notes, but not significantly
so. Q now about 0.68
Do these numbers sound even close to right???
What sort of data rate and filtering would your 32
bits need for the least significant bit not be lost in
the noise of the most significant bit? Is it a sensible
question, or I am dreaming a non-existent problem?
My little project needs 14bits (MIDI to Moog CV/Gate)
Arduino analog writes are only 8 bit PWM at 490Hz.
Dig down into the registers, maybe 12 bits...
So, the thought occurs to me: Use two PWMs and scale
them with resistors. Except the brain dead obvious 256x
ratio is so wide, its not easy to accurately dial in.
But MIDI control is only a pair of 7bits, so only 14bit
DAC is needed for those CV's. And 88 piano keys of
tuning is 186 DAC steps per key, 14bits seems plenty
there too...
So, I figure needing ony 7 bits PWM per analog write,
and a much more reasonable scaling factor of 64x. I
can switch back and forth between the redundant bits
that are within the range of either DAC to tweak 64x.
Now the tricky part. I need to low pass filter 490Hz,
and smooth the most significant bit enough that the
least significant bit isn't lost in the noise. -39dB.
I think a Sallen Key with 2 poles and a knee at 33Hz
can exceed that by a good amount, and 0.05 seconds
portamento to settle reasonably close to a new value.
Looking at 22K for both resistors, and 0.22uF for both
caps. Q is 0.5. I can't tell exact attenuation of 490Hz
from the crude graph, but sure looks to be more than
below the -40dB line...
64x scaling resistor is 1.5Meg in series with 68K.
I need 5K 25 turn pot (set to 2.5K) in series with
my Sallen's first 22K. I think that gives me enough
adjust to get away with worst case 5% resistors.
I'm never going to squeeze 7.33 octaves out of 5V
suppply rail at 1V per octave. But with 9V rail and
if I scale the Sallen's negative feedback, say 47K/
(22K+47K) I can stretch almost that far. Q rises,
but not so much its threatening to oscillate. And
gliding faster between notes, but not significantly
so. Q now about 0.68
Do these numbers sound even close to right???
What sort of data rate and filtering would your 32
bits need for the least significant bit not be lost in
the noise of the most significant bit? Is it a sensible
question, or I am dreaming a non-existent problem?
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.